Happy May!
I created new program and I want to test a idea that ODLS can be found directly in the output of the generators without first permuting them. This program checks dls about 40 times faster, but not as thoroughly. The same space is checked as batch 10, 12 and 13. Additionally the maximum number of diagonal transversals is recorded.
The workunit name is as follows:
padls4_a61923754 - application name
padls4_a61923754 - application compatibility version
padls4_a61923754 - generator version (a or b)
padls4_a61923754 - 8 numbers forming a input row for the padls generators

Credit will not be awarded during validation, but later during assimilation of the batch. Credit will be granted based on the number of Latin squares checked. If you do not agree with the experiment, the credit or the fact that your data is stored on my SD-Card, then please do not participate.

---

It was getting very disappointing that there were no ODLS, but now there are at least 24. So that confirms that there indeed are ODLS in plain generator output. There were errors and bunch of tasks had to be aborted, sorry for that, but this is experimental and mistakes can happen. The aborted tasks are now being resent with new app version. I also changed the server configuration so it will run more smoothly.

Rats, missed them all.

Waiting for the next batch.

Conan

Quickly looking at the incomplete results, only 14 ODLS were found. 28 after post-processing, all k=2. This is very disappointing result. Because we confirmed that there are ODLS in the raw output, but nowhere as many as with the family_mar program. Maybe the situation is different with the stronger generators.
But interesting is that there are duplicates! This means that the output of generators repeats in canonicial form. This was suspected, but not confirmed. I would like the output to not repeat, to avoid unnecessary repeated computation.
Miss Natalia developed more generator algorithms, but if they also exhibit this repeating behavior, then it's not the best way of finding ODLS. Ideally, the generators should give output that is unique after the transformation in family_mar, so no work is wasted.

The family_mar program generates about 380 squares from each input square and checks them at rate of 25 per second. But because it checks all the generated squares, it is actually checking them at rate of 9700 per second. This incredible speedup is caused by reusing some of the computation in checking a square for the whole family (hence the name). The simplified checker runs at about 1000 squares per second. If we can feed the family_mar program squares such that do not produce duplicates, then we can check 10 times faster! But I do not yet understand the permutations done by family_mar and much less how to generate input that does not repeat.

WUs where where there was square with more than 300 diagonal transversals: (wu-name max-transversals)

a39164725 328
a57932146 328
a57961243 328
a47632951 528
a61932754 538
a93174526 538
a41932756 574
a43179526 574
a67432951 574
a41723569 610
a93164527 610
a43169527 613
a67432159 613
a91723564 613

Interesting that no "b" is there. B units only had 199 transversals max, eg:

a27459163 199
a54137926 199
b35478269 199
b35487269 199
b38724965 199
b38924765 199
b45762983 199
b48679253 199
b48697253 199
b48697253 199
b54923786 199
b56823749 199
b58279346 199
b58297346 199
b58297346 199
b64278539 199
b64287539 199
b68752934 199
b94852736 199
b94852736 199
b95724863 199

All the ODLS found (none unique, all k=1):

LFLtNLGRqSPZSjHC3CjtmyZ6n #rule 38
L7U5H4AoYNWkLPNxeCinYdjVq #rule 38
SheXqCizRJXNHATv55fsEh2t3 #rule 51
Sskvz6eQ3N42foRb85c2F78sn #rule 51
SsPEjmEVxREjKvpzJ253ASmxG #rule 51
SwP8czj1d67VSfTjo8zxQknDF #rule 51
SyJb8hmtwADbJFpxF5Zm5tKSL #rule 51
T1AdB274yGz3KB9CA75jT3wgA #rule 51
T1AdB275BGypvtPmYRanPFVdY #rule 51
T2FE6Pqot7oBDjjCwHyKjZFQQ #rule 51
T2giRfhw73yimE5Fr6cxSv9p2 #rule 51
T4Nitz6YMMLm8MRHEGiVGxh4n #rule 51
T5o6zCdeeHoHncxsQ5zBSgurm #rule 51
T5yMRvchqMaG2k4ot7UARZ1TH #rule 51

count   	name	
2955	fzs600	
2238	Tomáš Brada	
729	LCB001	
277	Ralfy	
243	philip-in-hong kong	
125	Sebastien	
122	yoyo_rkn	

Thanks to the participants. They are sorted in this table by number of tasks done. I tried to assign credit, but it did not work. I however, WILL assign the credit at some point.

cav, i did not catch first batch. will wait for new )
its good you explain in details for all ,what hapening.. on other projects is just waiting for some response in days,,detto explain something from "back" scene))
good work

Very interesting! :)

Tomáš Brada
but did you know that exactly this work is performed by a BOINC project ODLK with 15, 38, 51 rulers (from March 24, 2018)?

Looks like you invented the bicycle :)

By the way, ODLS, which will be simply found from the DLS, produced by generators, will also be found when checking with the family_mar program.
But the family_mar program will find N times more solutions.

PS. In the meantime, I have several new experiments.
I regret that we did not agree on cooperation in your BOINC project.

I know that this run will be short so I did not spend much time to verify anything. I know about the other experiments (generator 4, two other rules, total generator ...). I can't wait to start a new experiment here. But, I must study now.

Credit assigned. 2.8932 per workunit.